Hofstadter talks about how the ability for computers to model the high level perception of the real world is not currently possible. He however says that using a microdomain is allowable. On page 190 he he says restricted domains can be the source of much insight. What I don’t understand is that Hoftstadter seems to not like the use of a computer program making an analogy between heat flow through a metal bar and water flow through a metal pipe. He says the fact that this program doesn’t know that water is wet and colorless questions the basis that it can claim an analogy was made. He states that there is an “insidious problem” talking about such a program.
He says that if the program doesn’t know every property of water then it can’t be used to make an analogy between water flow and heat flow. I don’t know if its soley the use of the word analogy that upsets Hoftstadter but it seems unfair to be againist what looks to me as a microdomain when he himself says its impractical for a true real world domain to be modeled.
I enjoyed the preface to chapter 5, more specifically the part Hofstadter talks about slips of the tongue or as he calls them word blends or substitution errors. These happen to everybody and isn’t seen as too big of a deal that we say something that makes no sense such as “Close your cookie jars, please” when referring to toy chests. This not only happens in speech, Hofstadter earlier in the book discussed this happening to him while typing.
This happens when the person is talking about 2 things that over lap in their separate conceptual halos. This is similar to what an analogy is. If something is overlapping with another something’s conceptual halos then those items are fairly similar and one could perhaps be used in place of another, accidently in the case of substitution errors. Hofstadter sees this concept as a basis to create a program dealing with analogies using the same architecture he used for Jumbo.
I enjoyed the section of the book that described the Eliza Effect. Which is when people read to much into what a computer displays or does and gives it more credit then it probably should.
For example the web site I use to host my blog, Tumblr, has a large image stating “Welcome Back” on its homepage. Thinking the site has any actual meaning behind displaying it would be falling to the Eliza Effect.
This can then be related to the feeling that the general public have that computers are smart. Hofstadter disagrees that a novel can be written by a computer. He says a computer can’t understand any word as well as a human can and by combining thousands of words into a novel is impossible for a computer to do.
The program described in the book “wrote” a book only because it had been programmed with thousands of rules and then would asked Scott French, writer of the program, questions on how to proceed. This hardly counts as writing a novel by itself but add on top of this that Scott French would change some words and correct some misspellings.
A computer given credit for writing a novel by looking up words in a database and asking a human questions, reminds me of the following saying. It has been said that a 1000 monkeys with 1000 typewriters given infinite time would eventually create the entire works of Shakespeare.
http://www.youtube.com/watch?v=iNNy3opa7Wc
At the end of chapter 3 Daniel Defays, creator of Numbo discusses the weaknesses of the architecture of Numbo. Numbo, to my understanding, only works towards a goal as it will create “blocks” which are the combination of “bricks” in order to get closer to the goal.
Daniel states that Numbo has some difficulty with some simple problems, one such problem is {Target = [41], Bricks = [5,16,22,25,1]}. The problem that Numbo has is that while it can build up to a number, it doesn’t know how to break apart a goal and work through that knowledge. For example 41 is very close to the number 40. 40 is a simpler number for a person to do math with, and a human knows that 40 is 20 plus 20, this is a fact that a human would probably use in the process of trying to solve this Numbo problem that the Numbo program doesn’t know. By knowing that 20 plus 20 is close to the goal and there being some bricks that are close to 20 create a few starting points for a human that Numbo wouldn’t have.
With this method of human thinking missing from Numbo it is understandable that Daniel thinks a “strict comparison of the performance of human subjects with the performance of Numbo is not possible”
Chapter 3 consists of Daniel Defays discussing his program Numbo which he worked with Hofstadter on. Numbo is similar to Jumbo, hence the name. Numbo is very similar to the crypto problems which we have been working on in class. The only difference between the two is a Numbo problem does not need to use all the bricks for it to be considered solved where a crypto problem does.
The reason for this is where a crypto problem is just a problem for a computer program to solve and a Numbo problem is used to study how a person thinks. If a person is given the problem of {Target = [87], Bricks = [8,3,9,10,7]} the rule of not having to use all 5 bricks is important in seeing how a person would solve the problem.
If all 5 bricks had to be used, the most common set of answers to the above problem which Daniel Defays says are “((8x10)+7)” and “((9x10)-3)” could not be accepted. While the problem {Target = [114], Bricks = [11,20,7,1,6]} and it’s most simplest solution of “((20-1)x6)” is almost never the solution a human comes up with as stated by Daniel.
The subject of when a human mind would come up with a longer solution then shorter one would not be available to discuss if the Numbo program only accepted answers of all its bricks being used to get to the target. In studying the human mind the simplist answer is one of the most important things that would be studied.
Throughout the second half of chapter 2, Hofstadter talks about how his Jumbo program reacts when its gone down the wrong path to a word and how it decides when to move backwards a step or two when it can’t go any farther and when to go back to the beginning.
Reading this I recalled when I was working on a programming assignment and was having difficulty finishing it. First I had first tired to make smaller adjustments to some methods that I thought were the things that were giving me problems. Once that proved unsuccessful after a few attempts, I then loaded an older version where I knew it worked but now would try to go about finishing the program by going down a different branch then I had the first time.
I can also relate Hofstadter’s idea of giving situations a temperature to my programming assignment. When I was having trouble with it, its temperature was warm due to things not working. A situation does not like being warm so I needed to backtrack and go back to when it was colder. The version I had decided to go back to was frozen as it was working, and since it was frozen, the happiest state for a situation, had prevented me from dissolving it and going back any further.
It is this method of backtracking that convinces me of Jumbo’s intelligence system is close to the way that a human would solve a Jumble puzzle.
On page 106 Hofstadter talks about the concept of terraced scans. He defines a terraced scan as a “parallel investigation of many possibilities to different levels of depth, quickly throwing out the bad ones and homing in rapidly and accurately on good ones.” He gives the example of a person looking for a book to read won’t start at the first book on the first shelf and read the whole book then move on to the book right next to it. They would go to a favorite genre and then narrow down further by reading a synopsis on the back of a book or deciding from the books of a favorite author.
He discusses this to explain why his Jumbo program does not include a dictionary. Instead of using a dictionary to narrow down what the solution to a jumble problem could be he uses the “preferedness” that parts of words have in being in a certain position of a word. He rates the string “sh” of having a rating of 8 for both the beginning of a word and the ending of a word. Therefore if there is an ’s’ and an ‘h’ in a Jumble problem there’s a good chance they are in the order of “sh” and it’s the start of the word or at the end of it.
He states that using his ”preferedness” method is more close to a human way of thinking than a dictionary method. Due to a person eliminating a disliked genre of books from even considering reading, Jumbo eliminates combinations that don’t prefer to be next to each other and where they are not often found in a word.
The part of the last reading that intrigued me the most was Hofstadter discussing the interesting things when words are split into “chunks”. If I saw a sign that read “N. Concord Rd.” nothing would have jumped out at me that anything very special was happening, besides the fact that I was driving on a road named North Concord. Hofstadter however immediately knew something interesting was hidden in the string “N. Concord Rd.”
It took him a minute but he was able to parse the interesting pattern “NCO-NCO-RD-RD” out of a seemingly ordinary street sign. He also mentions “hotshots” and “no nonsense” turn into “hots-hots” and “no-no-nse-nse” upon further inspection.
Just trying to think of words that this works for is very difficult. The best string I could come up with was “no nonsense hotshots”. Hofstadter and a student Henry Velick tried to help this by adapting the Jumbo program.
Hofstadter and Henry worked on a program that would attempt to take strings and attempt to find interesting chunks from them. They named this program Toreador. As with the term ”Seek-Whece” being a play on words I wonder if Toreador really means “To-read-or” or if it was only chosen because they liked the song from Carmen.
I found it interesting when Hoftstadter talks about “Conceptual Spheres”. At the center of a conceptual sphere is an event that happened and its surrounding layers are variants of this event. Over time this sphere will shrink and people’s memories will also decrease. He talks about why only the most similar things are affected by the event, be it interstate bridges, female tennis players, and drugs similar to Tylenol.
To further build on the bridge example he uses I’d like to discuss the collapse of I-35W Mississippi River Bridge in 2007.Much like the bridge in Connecticut that collapsed there were inspections of bridges following the collapse. This time it was only around 700 bridges with similar designs. Just like before the question can be raised, why only bridges like the one that collapsed were inspected, why not all bridges?
In the immediate period after this even when I had to drive over a bridge I would think of that bridge collapsing and wonder if it would happen to the bridge I’m currently driving over. 2 years after the tragedy when I cross over the bridge it never crosses my mind of how the bridge may collapse. This morning to get to class I had to drive over a bridge and I don’t believe I even realized I was crossing a bridge , I was probably more concerned with finding something to listen to on the radio.
On page 53 Hofstadter relates programs that figure out patterns to computers that play chess. He says neither really use the same methods that a human would in the same task. So I’d like to discuss how computers play chess and are capable of beating world champion chess players.
Computers that play chess are programmed to do what is called alpha-beta pruning. This is a method of eliminating things from being considered. It will examine a move and if by its calculations where at least one possibility of the move is worse than a previously examined move, it will stop evaluating this move.
While chess masters do things similar to this they have a few fundamental differences with computers that do this. Processors in computers can process these scenarios millions of times faster than a human can. Computers also have an easier time comparing whether a move is a good move or a bad move. A metric is given to each move which states how desirable a chess move is, where a move with a higher percentage of success is considered the best move. A human would not have as hard of a line separating a good play from a bad play that a computer would have, and human players would also have some feelings mixed in with their knowledge of chess determining which move to play. Another possible hindrance for human players is the fact that humans are not capable to forget something as easily as a computer can, which means the move is allows in the back of the players mind where a computer would have no problem in never thinking about it again.
